On the use of machine learning to identify topological rules in the packing of {beta}-strands

The machine learning program GOLEM was applied to discover topologicalrules in the packing ofß-sheets in /ß-domainproteins. Rules (constraints) were determined for four featuresof ß-sheet packing: (i) whether a ß-strandis at an edge; (ii) whether two consecutive ß-strandspack parallel or anti-parallel; (iii) whether twoß-strandspack adjacently; and (iv) the winding direction of two consecutiveß-strands. Rules were found with high predictive accuracyand coverage. The errors were generally associated with complicationsin domain folds, especially in one doubly wound domains. Investigationof the rules revealed interesting patterns, some of which wereknown previously, others that are novel. Novel features include(i) the relationship between pairs of sequential strands isin general one of decreasing size; (ii) more sequential pairsof strands wind in the direction out than in; and (iii) it takesa larger alteration in hydrophobicity to change a strand fromwinding in the direction out than in. These patterns in thedata may be the result of folding pathways in the domains. Therules found are of predictive value and could be used in thecombinatorial prediction of protein structure, or as a generaltest of model structures, e.g. those produced by threading.We conclude that machine learning has a useful role in the analysisof protein structures.